Name
Abstract | ||
|---|---|---|
The design of an optimal Bayesian classifier for multiple features is dependent on the estimation of multidimensional joint probability density functions and therefore requires a design sample size that increases exponentially with the number of dimensions. A method was developed that combines classification decisions from marginal density functions using an additional classifier. Unlike voting methods, this method can select a more appropriate class than the ones selected by the marginal classifiers, thus "overriding" their decisions. It is shown that this method always exhibits an asymptotic probability of error no worse than the probability of error of the best marginal classifier. The use of multiple features by a Bayesian classifier often leads to a reduced probability of error. Unfortunately, the design of an optimal Bayesian classifier for multiple features requires that the class-conditional probability density functions be known. If the density functions are not known a priori, they must be estimated from a design sample. The estimation of multidimensional joint probability density functions is often nontrivial and requires a design sample size that, in general, increases exponentially with the number of dimensions. This paper proposes a method for using multiple classifiers to obtain an acceptable probability of error from a design sample that is too small to permit an adequate estimate of the multivariate class-conditional density functions to be obtained. By using a supervisory Bayesian classifier to combine the classification decisions from a set of marginal Bayesian classifiers that each use only a subset of the features, an overall probability of error that is at least as good as that of the best of the marginal classifiers can be obtained. The theoretical performance of such a multiple classifier system is examined in detail below. |
Year | DOI | Venue |
|---|---|---|
2001 | 10.1142/S0218213001000465 | International Journal on Artificial Intelligence Tools |
Keywords | Field | DocType |
combination of classifiers,curse of dimensionality,pattern recognition,sample size,conditional probability,joint probability density function,bayesian classifier,probability of error | Joint probability distribution,Naive Bayes classifier,Pattern recognition,Computer science,Posterior probability,Artificial intelligence,Classifier (linguistics),Probabilistic classification,Margin classifier,Bayes classifier,Machine learning,Marginal distribution | Journal |
Volume | Issue | Citations |
10 | 1-2 | 0 |
PageRank | References | Authors |
0.34 | 7 | 2 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Mark D. Happel | 1 | 4 | 1.18 |
| Peter Bock | 2 | 0 | 0.34 |